Comparative Analysis of Wrapper, Filter, and Genetic Algorithm-Based Feature Selection Techniques on the Diagnostic Accuracy of Machine Learning Models for Diabetes Mellitus

• Joe Mary

  Author

Diabetes mellitus remains one of the most prevalent chronic diseases globally, with the International Diabetes Federation estimating that approximately 537 million adults currently live with the condition, a figure projected to reach 783 million by 2045 . Machine learning approaches have shown considerable promise in early diabetes detection; however, high-dimensional clinical datasets often contain irrelevant or redundant features that degrade classification performance, increase computational complexity, and hinder model interpretability . This study presents a comparative analysis of three feature selection paradigms—Filter methods, Wrapper methods, and Genetic Algorithm-based approaches—applied to the PIMA Indian Diabetes dataset to evaluate their impact on the diagnostic accuracy of machine learning models. Using Random Forest and XGBoost classifiers, the study employs a rigorous preprocessing pipeline incorporating missing value imputation, normalization, and SMOTE-based resampling to address class imbalance. The experimental results demonstrate that the Genetic Algorithm-based wrapper approach achieved the highest classification accuracy of 89.4%, outperforming traditional Filter methods (84.2%) and conventional Wrapper methods (86.7%). Feature importance analysis identified glucose, BMI, and age as the most influential predictors, consistent with prior clinical findings . The study contributes a replicable framework for feature selection optimization in diabetes prediction and provides actionable insights for healthcare practitioners seeking to implement accurate, interpretable diagnostic tools in primary care settings.

• PDF

Copyright (c) 2026 Joe Mary (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.